Effects of trimming weight-for-height data on growth-chart percentiles

Abstract

Before estimating smoothed percentiles of weight-for-height and BMI-for-age to construct the WHO growth charts, WHO excluded observations that were considered to represent unhealthy weights for height. The objective was to estimate the effects of similar data trimming on empirical percentiles from the CDC growth-chart data set relative to the smoothed WHO percentiles for ages 24-59 mo. We used the nationally representative US weight and height data from 1971 to 1994, which was the source data for the 2000 CDC growth charts. Trimming cutoffs were calculated on the basis of weight-for-height for 9722 children aged 24-71 mo. Empirical percentiles for 7315 children aged 24-59 mo were compared with the corresponding smoothed WHO percentiles. Before trimming, the mean empirical percentiles for weight-for-height in the CDC data set were higher than the corresponding smoothed WHO percentiles. After trimming, the mean empirical 95th and 97th percentiles of weight-for-height were lower than the WHO percentiles, and the proportion of children in the CDC data set above the WHO 95th percentile decreased from 7% to 5%. The findings were similar for BMI-for-age. However, for weight-for-age, which had not been trimmed by the WHO, the empirical percentiles before trimming agreed closely with the upper percentiles from the WHO charts. WHO data-trimming procedures may account for some of the differences between the WHO growth charts and the 2000 CDC growth charts.